pith. sign in

arxiv: 2604.22162 · v1 · submitted 2026-04-24 · 💻 cs.CV

SAMIDARE: Advanced Tracking-by-Segmentation for Dense Scenarios

Shozaburo Hirano , Norimichi Ukita This is my paper

Pith reviewed 2026-05-08 12:42 UTC · model grok-4.3

classification 💻 cs.CV
keywords multi-object trackingsports analysissegmentation-based trackingdense scenesocclusion handlingmask controlstate-aware associationSportsMOT
0
0 comments X p. Extension

The pith

SAMIDARE improves multi-object tracking in crowded sports scenes by regenerating masks adaptively and using state information to manage associations and new tracks.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper presents SAMIDARE as a way to fix common failures in segmentation-based tracking when many players overlap or leave the frame quickly. It adds three targeted changes to an existing base method: regenerating masks based on current density, updating memory only when useful, and linking tracks or starting new ones according to object states. This matters for sports video analysis because dense motion produces mask errors and identity flips that break downstream stats and highlights. If the changes work as described, they keep object features intact longer and cut down on switches without requiring a full new model.

Core claim

SAMIDARE enhances SAM2MOT for crowded scenes through density-aware mask re-generation and selective memory updates for adaptive mask control to preserve target feature integrity, along with state-aware association and new track initialization to improve robustness under mutual occlusions and frequent frame-out events. Evaluated on the SportsMOT dataset, it achieves state-of-the-art performance by outperforming the baseline by 2.5 HOTA and 4.2 IDF1 points on the validation set. These results show that adaptive feature management using mask control and state-aware association provides a robust and efficient solution for dense sports tracking.

What carries the argument

The SAMIDARE framework consisting of density-aware mask re-generation, selective memory updates for adaptive mask control, and state-aware association with new track initialization.

Load-bearing premise

The measured gains in tracking scores are produced by the three added components rather than other differences in code, training, or how the evaluation is run.

What would settle it

An ablation test on the SportsMOT validation set that turns off each of the three components one at a time and measures whether the 2.5 HOTA and 4.2 IDF1 advantages over the baseline disappear.

Figures

Figures reproduced from arXiv: 2604.22162 by Norimichi Ukita, Shozaburo Hirano.

Figure 1
Figure 1. Figure 1: Comparison in a dense basketball scenario. In view at source ↗
Figure 2
Figure 2. Figure 2: Contributions of our method. Here, only focus on the view at source ↗
Figure 3
Figure 3. Figure 3: Overview of the SAMIDARE pipeline. Our framework enhances mask propagation by integrating adaptive mask control (DA view at source ↗
Figure 4
Figure 4. Figure 4: Histogram of frame-out durations on the SportsMOT view at source ↗
Figure 5
Figure 5. Figure 5: Qualitative comparison in crowded sports scenarios in view at source ↗
read the original abstract

Automated sports analysis demands robust multi-object tracking (MOT), yet segmentation-based methods often struggle with mask errors and ID switches in dense scenes. We propose SAMIDARE, a framework that enhances SAM2MOT for crowded scenes through three key components: (1) density-aware mask re-generation and (2) selective memory updates, both for adaptive mask control to preserve target feature integrity, and (3) state-aware association and new track initialization, which improves robustness under mutual occlusions and frequent frame-out events. Evaluated on the SportsMOT dataset, SAMIDARE achieves state-of-the-art performance, outperforming the baseline by 2.5 HOTA and 4.2 IDF1 points on the validation set. These results demonstrate that adaptive feature management using mask control and state-aware association provide a robust and efficient solution for dense sports tracking. Code is available at https://github.com/ZabuZabuZabu/SAMIDARE

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 0 minor

Summary. The paper introduces SAMIDARE, an extension to SAM2MOT for multi-object tracking-by-segmentation in dense sports scenes. It proposes three components—density-aware mask re-generation, selective memory updates for adaptive mask control, and state-aware association with new track initialization—to handle mask errors, ID switches, mutual occlusions, and frame-out events. Evaluated on the SportsMOT dataset, the full model is reported to achieve state-of-the-art performance, outperforming the SAM2MOT baseline by 2.5 HOTA and 4.2 IDF1 points on the validation set. Code is released at the provided GitHub link.

Significance. If the reported gains can be rigorously attributed to the three components, the work would offer a targeted, practical improvement for segmentation-based MOT in crowded sports videos, an application area with clear downstream value. The public release of code is a clear strength that supports reproducibility and further analysis. At present, however, the lack of component isolation limits the ability to assess the precise contribution and generalizability of the approach.

major comments (1)
  1. [Abstract] Abstract: The central claim that SAMIDARE outperforms the SAM2MOT baseline by 2.5 HOTA and 4.2 IDF1 is presented solely as the full-model result on the SportsMOT validation set. No ablation studies, component-wise breakdowns, error analysis, or controlled re-implementations of the baseline are supplied, so it is impossible to verify that the gains arise from density-aware mask re-generation, selective memory updates, and state-aware association rather than from unstated implementation choices, hyper-parameter tuning, or evaluation-protocol details.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the thorough review and valuable suggestions. We address the major comment on the presentation of results and lack of ablations below. We agree that strengthening the evidence for component contributions will improve the paper.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The central claim that SAMIDARE outperforms the SAM2MOT baseline by 2.5 HOTA and 4.2 IDF1 is presented solely as the full-model result on the SportsMOT validation set. No ablation studies, component-wise breakdowns, error analysis, or controlled re-implementations of the baseline are supplied, so it is impossible to verify that the gains arise from density-aware mask re-generation, selective memory updates, and state-aware association rather than from unstated implementation choices, hyper-parameter tuning, or evaluation-protocol details.

    Authors: We appreciate the referee highlighting this issue. Upon review, the current manuscript indeed presents the performance gains primarily through the full model results in the abstract and evaluation section without dedicated ablation tables or breakdowns for each of the three components. While the paper explains the motivation and design of density-aware mask re-generation, selective memory updates for adaptive mask control, and state-aware association, it does not provide quantitative isolation of their individual impacts. We will revise the manuscript to include comprehensive ablation studies that evaluate the contribution of each component incrementally on the SportsMOT validation set. Additionally, we will provide error analysis and details on the baseline re-implementation to ensure transparency. These additions will allow for a clearer attribution of the 2.5 HOTA and 4.2 IDF1 improvements to the proposed methods rather than other factors. revision: yes

Circularity Check

0 steps flagged

No significant circularity in empirical claims or derivation

full rationale

The paper is an applied computer vision contribution that proposes three engineering components (density-aware mask re-generation, selective memory updates, state-aware association) and reports empirical gains on the public SportsMOT validation set against the external SAM2MOT baseline. No mathematical derivation chain, first-principles prediction, or fitted parameter is presented that reduces to its own inputs by construction. Performance numbers are measured on an external benchmark using standard HOTA/IDF1 metrics; the abstract and description contain no self-definitional equations, renamed known results, or load-bearing self-citations that would create circularity. The attribution of gains to the three components is an empirical question (addressed by the skeptic via lack of ablations), but that is a support/verification issue, not circularity.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the effectiveness of three new algorithmic components whose internal parameters and exact implementation are not specified in the abstract; relies on the domain assumption that SAM2MOT is a suitable baseline.

axioms (1)
  • domain assumption SAM2MOT serves as a valid and improvable baseline for segmentation-based multi-object tracking
    Paper builds directly on it without re-deriving or questioning its core validity.

pith-pipeline@v0.9.0 · 5463 in / 1229 out tokens · 42520 ms · 2026-05-08T12:42:28.631693+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

23 extracted references · 3 canonical work pages

  1. [1]

    Available from: https://arxiv.org/ abs/2206.14651

    Nir Aharon, Roy Orfaig, and Ben-Zion Bobrovsky. Bot- sort: Robust associations multi-pedestrian tracking.CoRR, abs/2206.14651, 2022. 2

    work page arXiv 2022

  2. [2]

    Observation-centric SORT: rethink- ing SORT for robust multi-object tracking

    Jinkun Cao, Jiangmiao Pang, Xinshuo Weng, Rawal Khi- rodkar, and Kris Kitani. Observation-centric SORT: rethink- ing SORT for robust multi-object tracking. InCVPR, pages 9686–9696, 2023. 1, 5

    2023

  3. [3]

    Price, Alexan- der G

    Ho Kei Cheng, Seoung Wug Oh, Brian L. Price, Alexan- der G. Schwing, and Joon-Young Lee. Tracking anything with decoupled video segmentation. InICCV, pages 1316– 1326, 2023. 5

    2023

  4. [4]

    Sportsmot: A large multi- object tracking dataset in multiple sports scenes

    Yutao Cui, Chenkai Zeng, Xiaoyu Zhao, Yichun Yang, Gangshan Wu, and Limin Wang. Sportsmot: A large multi- object tracking dataset in multiple sports scenes. InICCV, pages 9887–9897, 2023. 1, 5

    2023

  5. [5]

    Sam2long: Enhancing sam 2 for long video segmentation with a training-free memory tree

    Shuangrui Ding, Rui Qian, Xiaoyi Dong, Pan Zhang, Yuhang Zang, Yuhang Cao, Yuwei Guo, Dahua Lin, and Jiaqi Wang. Sam2long: Enhancing SAM 2 for long video segmentation with a training-free memory tree.CoRR, abs/2410.16268,

    work page arXiv

  6. [6]

    Strongsort: Make deep- sort great again.IEEE Trans

    Yunhao Du, Zhicheng Zhao, Yang Song, Yanyun Zhao, Fei Su, Tao Gong, and Hongying Meng. Strongsort: Make deep- sort great again.IEEE Trans. Multim., 25:8725–8737, 2023. 2

    2023

  7. [7]

    YOLOX: exceeding YOLO series in 2021.CoRR, 2021

    Zheng Ge, Songtao Liu, Feng Wang, Zeming Li, and Jian Sun. YOLOX: exceeding YOLO series in 2021.CoRR, 2021. 2, 5

    2021

  8. [8]

    Iterative scale-up expansioniou and deep features association for multi-object tracking in sports

    Hsiang-Wei Huang, Cheng-Yen Yang, Jiacheng Sun, Pyong- Kun Kim, Kwang-Ju Kim, Kyoungoh Lee, Chung-I Huang, and Jenq-Neng Hwang. Iterative scale-up expansioniou and deep features association for multi-object tracking in sports. InWACV Workshop, pages 163–172, 2024. 1, 2, 5

    2024

  9. [9]

    Sam2mot: A novel paradigm of multi-object tracking by segmentation

    Junjie Jiang, Zelin Wang, Manqi Zhao, Yin Li, and Dong- sheng Jiang. SAM2MOT: A novel paradigm of multi-object tracking by segmentation.CoRR, abs/2504.04519, 2025. 1, 2, 5

    work page arXiv 2025

  10. [10]

    Berg, Wan-Yen Lo, Piotr Doll ´ar, and Ross B

    Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chlo´e Rolland, Laura Gustafson, Tete Xiao, Spencer White- head, Alexander C. Berg, Wan-Yen Lo, Piotr Doll ´ar, and Ross B. Girshick. Segment anything. InICCV, pages 3992– 4003, 2023. 5

    2023

  11. [11]

    Matching anything by segmenting anything

    Siyuan Li, Lei Ke, Martin Danelljan, Luigi Piccinelli, Mattia Seg`u, Luc Van Gool, and Fisher Yu. Matching anything by segmenting anything. InCVPR, pages 18963–18973, 2024. 1, 5

    2024

  12. [12]

    Grounding DINO: mar- rying DINO with grounded pre-training for open-set object detection

    Shilong Liu, Zhaoyang Zeng, Tianhe Ren, Feng Li, Hao Zhang, Jie Yang, Qing Jiang, Chunyuan Li, Jianwei Yang, Hang Su, Jun Zhu, and Lei Zhang. Grounding DINO: mar- rying DINO with grounded pre-training for open-set object detection. InECCV, pages 38–55, 2024. 5

    2024

  13. [13]

    Jonathon Luiten, Aljosa Osep, Patrick Dendorfer, Philip H. S. Torr, Andreas Geiger, Laura Leal-Taix ´e, and Bastian Leibe. HOTA: A higher order metric for evaluating multi- object tracking.Int. J. Comput. Vis., 129(2):548–578, 2021. 5

    2021

  14. [14]

    Diffmot: A real-time diffusion-based multiple object tracker with non-linear prediction

    Weiyi Lv, Yuhang Huang, Ning Zhang, Ruei-Sung Lin, Mei Han, and Dan Zeng. Diffmot: A real-time diffusion-based multiple object tracker with non-linear prediction. InCVPR, pages 19321–19330, 2024. 5

    2024

  15. [15]

    Motiontrack: Learning robust short-term and long-term motions for multi-object tracking

    Zheng Qin, Sanping Zhou, Le Wang, Jinghai Duan, Gang Hua, and Wei Tang. Motiontrack: Learning robust short-term and long-term motions for multi-object tracking. InCVPR, pages 17939–17948, 2023. 2

    2023

  16. [16]

    Towards generalizable multi-object tracking

    Zheng Qin, Le Wang, Sanping Zhou, Panpan Fu, Gang Hua, and Wei Tang. Towards generalizable multi-object tracking. InCVPR, pages 18995–19004, 2024. 5

    2024

  17. [17]

    Girshick, Piotr Doll ´ar, and Christoph Feichtenhofer

    Nikhila Ravi, Valentin Gabeur, Yuan-Ting Hu, Ronghang Hu, Chaitanya Ryali, Tengyu Ma, Haitham Khedr, Roman R¨adle, Chlo ´e Rolland, Laura Gustafson, Eric Mintun, Junt- ing Pan, Kalyan Vasudev Alwala, Nicolas Carion, Chao- Yuan Wu, Ross B. Girshick, Piotr Doll ´ar, and Christoph Feichtenhofer. Segment anything model 2.1.https: //github.com/facebookresearch...

    2024

  18. [18]

    Girshick, Piotr Doll ´ar, and Christoph Feichtenhofer

    Nikhila Ravi, Valentin Gabeur, Yuan-Ting Hu, Ronghang Hu, Chaitanya Ryali, Tengyu Ma, Haitham Khedr, Roman R¨adle, Chlo ´e Rolland, Laura Gustafson, Eric Mintun, Junt- ing Pan, Kalyan Vasudev Alwala, Nicolas Carion, Chao- Yuan Wu, Ross B. Girshick, Piotr Doll ´ar, and Christoph Feichtenhofer. SAM 2: Segment anything in images and videos. InICLR, 2025. 1, 2

    2025

  19. [19]

    Zou, Rita Cuc- chiara, and Carlo Tomasi

    Ergys Ristani, Francesco Solera, Roger S. Zou, Rita Cuc- chiara, and Carlo Tomasi. Performance measures and a data set for multi-target, multi-camera tracking. InECCV, pages 17–35, 2016. 5

    2016

  20. [20]

    No train yet gain: Towards generic multi-object tracking in sports and beyond

    Tomasz Stanczyk, Seongro Yoon, and Franc ¸ois Br ´emond. No train yet gain: Towards generic multi-object tracking in sports and beyond. InCVPRW, pages 6039–6048, 2025. 5

    2025

  21. [21]

    Raymond, and Pritam Chanda

    Balaji Veeramani, John W. Raymond, and Pritam Chanda. Deepsort: deep convolutional networks for sorting haploid maize seeds.BMC Bioinform., 19-S(9):85–93, 2018. 2

    2018

  22. [22]

    SAMURAI: motion- aware memory for training-free visual object tracking with SAM 2.IEEE Trans

    Cheng-Yeng Yang, Hsiang-Wei Huang, Wenhao Chai, Zhongyu Jiang, and Jenq-Neng Hwang. SAMURAI: motion- aware memory for training-free visual object tracking with SAM 2.IEEE Trans. Image Process., 35:970–982, 2026. 1

    2026

  23. [23]

    Bytetrack: Multi-object tracking by associating every detection box

    Yifu Zhang, Peize Sun, Yi Jiang, Dongdong Yu, Fucheng Weng, Zehuan Yuan, Ping Luo, Wenyu Liu, and Xinggang Wang. Bytetrack: Multi-object tracking by associating every detection box. InECCV, pages 1–21, 2022. 1, 2, 5

    2022